A High-Resolution Two-Stage Satellite Model to Estimate PM2.5 Concentrations in China

Abstract:

With the rapid economic development and urbanization, severe and widespread PM2.5 pollution in China has attracted nationwide attention. Study of the health impact of PM_2.5 exposure has been hindered, however, by the limited coverage of ground measurements from recently established regulatory monitoring networks. Estimating ground-level PM_2.5 from satellite remote sensing is a promising new method to evaluate the spatial and temporal patterns of PM_2.5 exposure. We developed a two-stage spatial statistical model to estimate daily mean PM2.5 concentrations at 10 km resolution in 2013 in China using MODIS Collection 6 AOD, assimilated meteorology, population density, and land use parameters. A custom inverse variance weighting approach was developed to combine MODIS Dark Target (DT) and Deep Blue (DB) AOD to optimize coverage. Compared with the AERONET AOD measurements, our combined AOD (R²=0.80, mean bias = 0.07) performs similarly to MODIS’ combined AOD (R²=0.81, mean bias =0.07), but has 90% greater coverage. We used the first-stage linear mixed effect model to represent the temporal variability of PM_2.5 and the second-stage generalized additive model to represent its spatial contrast. The overall model cross-validation R² and relative prediction error are 0.80 and 30%, respectively. PM_2.5 levels exhibit strong seasonal patterns, with the highest national mean concentrations in winter (75 µg/m³) and the lowest in summer (30 µg/m³). Elevated annual mean PM_2.5 levels are predicted in North China Plain and Sichuan Basin, with the maximum annual PM_2.5 concentrations higher than 130 µg/m³ and 110 µg/m³, respectively. Our results also indicates that over 94% of the Chinese population lives in areas that exceed the WHO Air Quality Interim Target-1 standard (35 μg/m³). The exceptions include Taiwan, Hainan, Yunnan, Tibet, and North Inner Mongolia.